After call to REngine.GetInstance, app's own Ctrl+C handler no longer works

[kierenj]

I'm using .NET Core - as stated, after I've called GetInstance, my Ctrl+C handler (added via Console.CancelKeyPress += ... is no longer called when I hit Ctrl+C

[lrasmus]

@kierenj - is there a minimal set of code (or your repository) you can provide to reproduce the problem? Just want to make sure we're able to reproduce the problem exactly.

[kierenj]

No problem:

https://github.com/kierenj/rdotnet-ctrl-c-repro

master branch ignores the CancelKeyPressHandler working branch works fine - because the call to REngine.GetInstance is commented out

Let me know if I can assist/help any further

[lrasmus]

I am by no means an expert, but this may be "working as designed". It appears to be more of the way embedded R works than R.NET (at least from what I can see). When the REngine is initialized, and it's in turn initializing R it starts up R's REPL: https://github.com/rdotnet/rdotnet/blob/master/R.NET/REngine.cs#L493

Which eventually gets to the REPL code which I believe is first grabbing your keyboard input and no longer passing it up to your handler: https://github.com/wch/r-source/blob/32eff2cd469115952a6f19eccc9c0b08c8d62822/src/main/main.c#L200-L295

The reason I'm thinking this may be as designed is that when I'm in the R console natively and I press Ctrl+C, it takes the input although it does no action with it.

I tried a few things to get around it. First, REngine's Initialize takes a character device as a parameter. If one isn't specified (default behavior), it creates one. I tried using the NullCharacterDevice explicitly, creating my own character device just to see what callbacks I could catch, and skipping over that initialization code within REngine. Unfortunately, none of those made a difference. I think it's because that callbacks happen way after R has grabbed the input.

[kierenj]

ah - it looks like this one - https://github.com/wch/r-source/blob/32eff2cd469115952a6f19eccc9c0b08c8d62822/src/main/main.c#L671 called by https://github.com/wch/r-source/blob/32eff2cd469115952a6f19eccc9c0b08c8d62822/src/main/main.c#L911

That's checking R_SignalHandlers which is hard-set here: https://github.com/wch/r-source/blob/32eff2cd469115952a6f19eccc9c0b08c8d62822/src/main/main.c#L708

I'm very new to this too... does that mean I'd need a custom build of R?

[lrasmus]

@kierenj - sorry for not closing the loop on this before. Yes, I think it would mean a custom build of R. I've been trying to actually do this to verify it'll solve the problem, but have just had some problems getting R to build in my environment. I'm not sure if you had progressed with this, but curious to know where you stand with this issue.

[kierenj]

Our project for the moment has drawn to a close, and it remains as just an inconvenience on Windows. When deployed (docker/Linux), it responds to SIGTERM just fine. It's a bit of a pain but not the end of the world.